Using transcription-based detectors to emulate the behaviour of sequential probability ratio-based concentration detectors

TL;DR

This paper proposes a transcription-based detection method that emulates the hit rate of the sequential probability ratio test (SPRT) in cellular decision-making, achieving similar speed and accuracy without computing the log-likelihood ratio.

Contribution

It introduces a novel transcription-based detector that mimics SPRT's behavior using promoter binding rates, bypassing the need for explicit likelihood ratio calculations.

Findings

Transcription-based detector can approximate SPRT hit rate.

The detector's detection time is comparable or better than SPRT.

It operates effectively over a range of concentrations.

Abstract

The sequential probability ratio test (SPRT) from statistics is known to have the least mean decision time compared to other sequential or fixed-time tests for given error rates. In some circumstances, cells need to make decisions accurately and quickly, therefore it has been suggested the SPRT may be used to understand the speed-accuracy tradeoff in cellular decision making. It is generally thought that in order for cells to make use of the SPRT, it is necessary to find biochemical circuits that can compute the log-likelihood ratio needed for the SPRT. However, this paper takes a different approach. We recognise that the high-level behaviour of the SPRT is defined by its positive detection or hit rate, and the computation of the log-likelihood ratio is just one way to realise this behaviour. In this paper, we will present a method which uses a transcription-based detector to emulate the hit rate of the SPRT without computing the exact log-likelihood ratio. We consider the problem of using a promoter with multiple binding sites to accurately and quickly detect whether the concentration of a transcription factor is above a target level. We show that it is possible to find binding and unbinding rates of the transcription factor to the promoter's binding sites so that the probability that the amount of mRNA produced will be higher than a threshold is approximately equal to the hit rate of the SPRT detector. Moreover, we show that the average time that this transcription-based detector needs to make a positive detection is less than or equal to that of the SPRT for a wide range of concentrations. We remark that the last statement does not contradict Wald's optimality result because our transcription-based detector uses an open-ended test.